“Minwon_Itda”: An AI-Based Personalized Public Service Guidance System Using Hybrid Retrieval and Retrieval-Augmented Generation

Accessing public administrative services often requires citizens to navigate complex procedures, interpret specialized terminology, and identify eligibility requirements and supporting documents. Conventional e-government portals primarily rely on keyword-based search and predefined categories, which can make service discovery difficult for users who are unfamiliar with administrative terminology. This paper presents Minwon Itda, an AI-based personalized public service guidance system designed to facilitate the discovery and completion of administrative procedures through natural-language interaction. The proposed system integrates a two-stage service classification mechanism combining rule-based keyword mapping with a large language model (LLM), together with a hybrid retrieval pipeline that combines semantic vector search and keyword matching. Retrieved information is used to generate context-grounded responses through retrieval-augmented generation (RAG). Beyond conversational question answering, the system provides structured, step-by-step procedural guidance, personalized document checklists, administrative terminology explanations, and multilingual interaction. A session-management mechanism maintains users’ application progress, while caching and fallback mechanisms support responsive and resilient service delivery. The system was implemented using a web-based three-tier architecture with a FastAPI backend, a Next.js frontend, MariaDB, and Redis. The reported evaluation achieved a public-service classification accuracy of 98%. In latency measurements conducted over 100 requests, the average response time was approximately 0.4 s for keyword-matched queries and 1.8 s when LLM augmentation was invoked, while cached responses were delivered in less than 0.1 s on average. These results demonstrate the technical feasibility of integrating hybrid retrieval, LLM-based response generation, and structured procedural guidance into a unified public-service application. The proposed system provides an implementation framework for improving access to administrative information, although further user studies and broader evaluations are required to establish its effectiveness across diverse populations and service scenarios.

E-Government, Public Service Chatbot, Personalized Service Guidance, Large Language Models, Retrieval-Augmented Generation, Hybrid Information Retrieval

1. M. E. Cortés-Cediel, A. Segura-Tinoco, I. Cantador, and M. P. Rodríguez Bolívar, “Trends and challenges of e-government chatbots: Advances in exploring open government data and citizen participation content,” Government Information Quarterly, vol. 40, no. 4, Art. no. 101877, 2023. [Google Scholar] [Crossref]

2. A. G. Larsen and A. Følstad, “The impact of chatbots on public service provision: A qualitative interview study with citizens and public service providers,” Government Information Quarterly, vol. 41, no. 2, Art. no. 101927, 2024. [Google Scholar] [Crossref]

3. N. Aoki, “An experimental study of public trust in AI chatbots in the public sector,” Government Information Quarterly, vol. 37, no. 4, Art. no. 101490, 2020. [Google Scholar] [Crossref]

4. E. Adamopoulou and L. Moussiades, “An overview of chatbot technology,” in Artificial Intelligence Applications and Innovations (AIAI 2020), pp. 373–383. Springer, 2020. [Google Scholar] [Crossref]

5. A. Følstad and P. B. Brandtzaeg, “Chatbots and the new world of HCI,” Interactions, vol. 24, no. 4, pp. 38–42, 2017. [Google Scholar] [Crossref]

6. P. Lewis et al., “Retrieval-augmented generation for knowledge-intensive NLP tasks,” in Advances in Neural Information Processing Systems, vol. 33, 2020. [Google Scholar] [Crossref]

7. Y. Gao et al., “Retrieval-augmented generation for large language models: A survey,” arXiv preprint arXiv:2312.10997, 2023. [Google Scholar] [Crossref]

8. V. Karpukhin et al., “Dense passage retrieval for open-domain question answering,” in Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pp. 6769–6781, 2020. [Google Scholar] [Crossref]

9. N. Thakur, N. Reimers, A. Rücklé, A. Srivastava, and I. Gurevych, “BEIR: A heterogeneous benchmark for zero-shot evaluation of information retrieval models,” in Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks, vol. 1, 2021. [Google Scholar] [Crossref]

10. R. Nogueira and K. Cho, “Passage re-ranking with BERT,” arXiv preprint arXiv:1901.04085, 2019. [Google Scholar] [Crossref]

• Smart Safety Monitoring for Construction Sites